Superpowers
obra/superpowersA 15-skill development methodology that makes coding agents clarify the goal, write a plan, work through it with subagents and verify before calling it done.
Recommended
Check first Depth 3/5
wshobson/agents Claude skills
A multi-harness plugin marketplace with 94 plugins, 202 agents, 185 skills and 105 commands for Claude Code, Codex, Cursor, OpenCode, Copilot, Antigravity and Pi.
Our verdict Recommended with caveats
A widely used, MIT-licensed catalog with good breadth, but skill depth varies, evaluation is thin and you should install plugins selectively.
What wshobson/agents does
This is a large catalog of plugins that bundle agents, skills and slash commands by job: Python and JavaScript/TypeScript development, developer essentials (code review, debugging, Git, testing), security scanning, accessibility compliance, backend architecture patterns, infrastructure and many more. The README lists 94 plugins with 202 agents, 185 skills and 105 commands, kept in one Markdown source and adapted per tool. Claude Code and Cursor use plugin marketplaces; Codex has a native marketplace route plus generated adapters; OpenCode, Antigravity, GitHub Copilot and Pi are installed from a clone with make targets. Individual skills can also be installed with gh skill or npx skills.
The skills are short, about 5.4 KB on average, and usually add one or two reference files; examples are WCAG audit patterns, screen reader testing, CQRS and event-store design, FastAPI templates, parallel debugging and team-composition patterns for multi-agent work. A docs folder covers a capability matrix per harness, and the README says that its LLM-judge and Monte Carlo evaluation layers are experimental and not validated against human labels; the repo has two eval files. The practical approach is to install the few plugins that match your stack, not the whole marketplace.
Good for
- Developers who want ready-made agents and skills for a mainstream stack
- Teams using more than one coding agent who want a shared source
- Picking individual skills such as accessibility or API design patterns
Not for
- Anyone expecting every plugin to be deep; most skills are short pattern guides
- Installing the whole catalog at once, which adds a lot of agents and commands to manage
What's inside depth 2/5
185 skills averaging about 5.4 KB with one or two reference files in many; only a few templates or scripts and two eval files for a catalog this size.
- Skills
- 185
- Reference docs
- 244
- Scripts
- 8
- Tests / evals
- 2
- Templates
- 15
- Avg SKILL.md
- 5.3 KB
Counted from the repository file tree. How we score depth.
Dependencies and costs
- A supported coding agent
- GitHub CLI 2.90+ for gh skill, or Node.js for npx skills
- make for the clone-based installers
Security review
Review before installingA very large catalogue of Markdown agent and skill definitions grouped into plugins. The files we have are documentation and templates; the exceptions are brand-landingpage, which has the agent read the online Stitch SDK documentation to install and call the SDK, and block-no-verify, whose global setup would overwrite ~/.claude/settings.json. Plugin manifests, hooks and installers were not in the reviewed copy, so read the plugin you install.
Worth knowing
- Installs packages or toolsIf the Stitch SDK is missing the agent is told to install it, globally by default, and to check a Stitch API key in the environment (never echoing it) before making SDK calls.plugins/brand-landingpage/skills/brand-landingpage/SKILL.md:32
- Changes files outside your projectThe block-no-verify guide adds a PreToolUse hook to .claude/settings.json and gives a 'Global Setup' of cat > ~/.claude/settings.json (a heredoc that would replace the whole global settings file); the README install paths (gh skill, npx skills, make install-*) also write into agent config folders.plugins/block-no-verify/skills/block-no-verify-hook/SKILL.md:114
- Runs remote code at an unpinned versionInstall instructions use unpinned npx skills add wshobson/agents --skill <name> and gh skill install; several skills' reference files also show unpinned npx @axe-core/cli, npx pa11y and pip install snippets as examples.README.md:76
- Relies on a CLI or package we could not reviewRelies on GitHub CLI 2.90+, Node/npm, uv/Python and make for the clone-based installers (make install-opencode etc.); the installer scripts and Makefile are not in the reviewed files.README.md:76
- Makes network requestsThe brand-landingpage skill sends your brand interview answers and design prompts to the Stitch service through its SDK/MCP tools; most other skills are reference Markdown with example curl commands against placeholder hosts.plugins/brand-landingpage/skills/brand-landingpage/SKILL.md:167
What we checked
The repository (1170 files) on Oct 29, 2026. Not a full review: 1073 of 1170 files were not downloaded (large repository; we reviewed the skill folders); it relies on code outside the repository: Stitch SDK and Stitch service (stitch-design.ai), and the online documentation the brand-landingpage skill follows; Makefile and installer scripts for OpenCode, Antigravity, Copilot and Pi, plus the marketplace manifests and plugin.json files for Claude, Codex and Cursor (about 1,073 readable files, including .agents, .cursor-plugin and .claude-plugin folders, are not in the local copy) so plugin-level hooks, MCP configs and agent definitions could not be checked; GitHub CLI (gh skill), the npx skills installer, and the npm/pip tools shown in example commands.
NVIDIA SkillSpector static scan: 45 pattern hits (MCP Rug Pull 11, Privilege Escalation 9, Tool Misuse 8, Data Exfiltration 7, Agent Snooping 6, Excessive Agency 3). The 45 hits are documentation: .env and 'access token' lines in a FastAPI template guide, kubeconfig and hostPath in CI and Kubernetes examples, DELETE /api/users/{id} in a REST guide, git --no-verify strings inside the block-no-verify hook guide (which blocks them), and unpinned npx examples.
A review, not a certification. Skills change after we check them — read the files before installing.
License
MIT — commercial use allowed. External plugins listed in the catalog have their own licenses.
Can it run unattended?
Not really. Individual plugins vary; the multi-agent coordination skills describe team patterns but the catalog as a whole has no unattended-run guarantees.
Skills in this repository
| Skill | Installs |
|---|---|
| typescript-advanced-types | 94K |
| tailwind-design-system | 71K |
Linked skills have their own review.
Alternatives to wshobson/agents
Claude Skills (alirezarezvani)
alirezarezvani/claude-skillsA 400+ skill library of agents, personas and commands for engineering, marketing, product, compliance and C-level work, mirrored for 13 coding agents.
With caveats
Check first Depth 3/5
Full-Stack Dev Skills (Jeffallan)
Jeffallan/claude-skills67 role-based developer skills for languages, frameworks, DevOps and security, each with reference files, plus 9 Jira/Confluence workflow commands.
Recommended
Check first Depth 3/5
ECC (Everything Claude Code)
affaan-m/ECCA very large Claude Code harness: 311 skills, 71 agents, commands, hooks, memory, rules and a scanner, with adapters for Codex, Cursor and others.
With caveats
Broad access Depth 3/5
FAQ
What is wshobson/agents?
It is a marketplace of 94 plugins with 202 agents, 185 skills and 105 commands for coding agents. The same Markdown source is adapted for Claude Code, Codex, Cursor, OpenCode, GitHub Copilot, Antigravity and Pi.
How do I install wshobson/agents in Claude Code?
Run /plugin marketplace add wshobson/agents and then install a plugin, for example /plugin install python-development@claude-code-workflows. For single skills you can use npx skills add wshobson/agents --skill <name>.
Is wshobson/agents safe?
It is an MIT-licensed catalog of markdown agents, skills and commands, and we found no obfuscated code. Plugins may include hooks or MCP settings, so check what each plugin adds before enabling it.
wshobson/agents vs Superpowers: which should I use?
wshobson/agents is a broad catalog of domain agents and pattern guides you pick from. Superpowers is a smaller, opinionated process pack (plan, subagents, TDD, verification), so choose it if you want one workflow enforced.
Sources: GitHub repository (stars, license, file tree, last push Oct 26, 2026); install counts from skills.sh via Best Skills (CC BY 4.0). Review based on the README and SKILL.md files as published.